MSN: Rethinking Cross-Domain Automatic Modulation Recognition via Decoupled Modulation Signal Normalization
IEEE Transactions on Cognitive Communications and Networking · 2026 · 12 · 8492-8505
DOI: 10.1109/tccn.2026.3695841 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Cognitive Communications and Networking, 2026. Verfasst von Yifan Chen und 4 weiteren Autor:innen. Die Arbeit nennt 49 Literaturangaben. Gefördert von: Fundamental Research Funds for the Central Universities, Natural Science Basic Research Program of Shaanxi Province, China Postdoctoral Science Foundation.
Abstract (Originalsprache)
Normalization is known to enhance the generalizability of deep neural networks. Curiously, most normalization methods are designed for images, language, and time-series data, without considering the unique characteristics of communication signals. Due to joint impairments from the entire transmission-propagation-reception chain, existing normalizations cannot be directly adapted to Automatic Modulation Recognition (AMR). To this end, this paper rethinks normalization in the context of AMR and uncovers the underlying factors that determine its effectiveness. Specifically, it first identifies that distribution shifts in AMR hinder model performance and systematically classifies distribution shifts in AMR into two types: (1) intra-chain shift, caused by temporal variations due to various physical effects, and (2) inter-chain shift, resulting from spatial variations arising from inherent environmental factors. From this characterization, we argue that intra-chain shifts manifest as temporal variations of signals, while inter-chain shifts introduce instance-specific spatial distortions. Motivated by this insight, this work proposes Modulation Signal Normalization (MSN), a decoupled normalization mechanism. MSN mitigates intra-chain shifts by smoothing feature correlations to extract stationary temporal components. Simultaneously, it extracts instance-specific statistical information and localized spatial details as stationary spatial components, which serve as discriminative cues for inter-chain shifts. This targeted design effectively addresses both types of shifts, thereby enhancing the cross-domain robustness of AMR models. Finally, extensive experiments demonstrate that MSN consistently enhances the generalizability of AMR models, achieving outstanding results.
Förderung
- Fundamental Research Funds for the Central Universities
- Natural Science Basic Research Program of Shaanxi Province
- China Postdoctoral Science Foundation
- Shaanxi Province Innovation Capability Support Plan
- National Natural Science Foundation of China
Zitieren
APA
Chen, Y., Zhang, X., Wang, G., Liu, Z., & Jiao, L. (2026). MSN: Rethinking Cross-Domain Automatic Modulation Recognition via Decoupled Modulation Signal Normalization. IEEE Transactions on Cognitive Communications and Networking, 12, 8492-8505. https://doi.org/10.1109/tccn.2026.3695841
Vancouver
Chen Y, Zhang X, Wang G, Liu Z, Jiao L. MSN: Rethinking Cross-Domain Automatic Modulation Recognition via Decoupled Modulation Signal Normalization. IEEE Transactions on Cognitive Communications and Networking. 2026;12:8492-8505. doi:10.1109/tccn.2026.3695841
BibTeX
@article{Chen2026,
author = {Chen, Yifan and Zhang, Xiangrong and Wang, Guanchun and Liu, Ziyi and Jiao, Licheng},
title = {MSN: Rethinking Cross-Domain Automatic Modulation Recognition via Decoupled Modulation Signal Normalization},
journal = {IEEE Transactions on Cognitive Communications and Networking},
year = {2026},
volume = {12},
pages = {8492--8505},
doi = {10.1109/tccn.2026.3695841}
}
EUVIMED